| 1. |
- Rosén, Emil
(författare)
-
Modeling glioblastoma growth patterns and their mechanistic origins
- 2021
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Glioblastoma (GBM) is the most common and aggressive primary brain cancer. GBM cells migrate away from the primary lesion and invade healthy brain tissue. The invading cells escape surgical resection, radiotherapy and develop resistance to chemotherapy. Consequently, despite treatment, recurrence is inevitable, and survival is only 14 months. For this purpose, we conducted four studies where we integrated experimental data from extensive patient material with image analysis and mathematical modeling.In study 1, we developed a tool, TargetTranslator, integrating different data modalities to identify new treatments. We implemented an image analysis pipeline to validate our results using a deep artificial neural network to quantify neuroblastoma cell differentiation.In study 2, we integrated the zebrafish and image analysis from study 1 to develop a high-throughput in vivo assay. Zebrafish were orthotopically injected with GBM cells, and each fish's tumor growth and vital status were automatically measured. We characterized the in vivo proliferation rate, survival, and treatment response to the drug marizomib for several patient-derived cell cultures. Light-sheet imaging also revealed two distinct growth types. The first set of cell cultures grew as bulk tumors, whereas the second set invaded vasculature as single cells.In study 3, we used the image analysis from study 1, coupled with an agent-based model to estimate in vitro cell migration and proliferation from single end-point images. The method was validated by a time series data set and applied to a large high-content drug screen of GBM cells. We identified three promising candidates for reducing GBM cell migration. The method can estimate migration on any end-point images of adherent cells without any additional experimental cost.Study 4 characterized the growth and invasive patterns of 45 patient-derived GBM cell cultures in orthogonal mouse xenografts. We found that up to four independent axes of variation could describe the phenotypes and were associated with distinct transcriptomic pathways. The transcriptomic pathways were in part associated with common genomic alterations and subtypes in GBM. We further identified a particularly aggressive GBM phenotype.In conclusion, this thesis was interdisciplinary and aimed to measure survival, invasion, and morphology from extensive patient material. The work had given us new insight into GBM invasion and growth and developed several scalable models suitable for evaluating new therapies.
|
|
| 2. |
- Fries, Niklas, 1991-
(författare)
-
Data-driven quality management using explainable machine learning and adaptive control limits
- 2023
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- In industrial applications, the objective of statistical quality management is to achieve quality guarantees through the efficient and effective application of statistical methods. Historically, quality management has been characterized by a systematic monitoring of critical quality characteristics, accompanied by manual and experience-based root cause analysis in case of an observed decline in quality. Machine learning researchers have suggested that recent improvements in digitization, including sensor technology, computational power, and algorithmic developments, should enable more systematic approaches to root cause analysis.In this thesis, we explore the potential of data-driven approaches to quality management. This exploration is performed with consideration to an envisioned end product which consists of an automated data collection and curation system, a predictive and explanatory model trained on historical process and quality data, and an automated alarm system that predicts a decline in quality and suggests worthwhile interventions. The research questions investigated in this thesis relate to which statistical methods are relevant for the implementation of the product, how their reliability can be assessed, and whether there are knowledge gaps that prevent this implementation.This thesis consists of four papers: In Paper I, we simulated various types of process-like data in order to investigate how several dataset properties affect the choice of methods for quality prediction. These properties include the number of predictors, their distribution and correlation structure, and their relationships with the response. In Paper II, we reused the simulation method from Paper I to simulate multiple types of datasets, and used them to compare local explanation methods by evaluating them against a ground truth.In Paper III, we outlined a framework for an automated process adjustment system based on a predictive and explanatory model trained on historical data. Next, given a relative cost between reduced quality and process adjustments, we described a method for searching for a worthwhile adjustment policy. Several simulation experiments were performed to demonstrate how to evaluate such a policy.In Paper IV, we described three ways to evaluate local explanation methods on real-world data, where no ground truth is available for comparison. Additionally, we described four methods for decorrelation and dimension reduction, and describe the respective tradeoffs. These methods were evaluated on real-world process and quality data from the paint shop of the Volvo Trucks cab factory in Umeå, Sweden.During the work on this thesis, two significant knowledge gaps were identified: The first gap is a lack of best practices for data collection and quality control, preprocessing, and model selection. The other gap is that although there are many promising leads for how to explain the predictions of machine learning models, there is still an absence of generally accepted definitions for what constitutes an explanation, and a lack of methods for evaluating the reliability of such explanations.
|
|
| 3. |
- Johansson, Patrik
(författare)
-
Large scale integration and interactive exploration of cancer data – with applications to glioblastoma
- 2018
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Glioblastoma is the most common malignant brain tumor, with a median survival of approximately 15 months. The standard of care treatment consists of surgical resection followed by radiotherapy and chemotherapy, where chemotherapy only prolongs survival by approximately 3 months. There is therefore an urgent need for new approaches to better understand the molecular vulnerabilities of glioblastoma. To this end, we have conducted four interdisciplinary studies.In study 1 we develop a method for efficiently constructing and exploring large integrative network models that include multiple cohorts and multiple types of molecular data. We apply this method to 8 cancers from The Cancer Genome Atlas (TCGA) and make the integrative network available for exploration and visualization through a custom web interface.In study 2 we establish a biobank of 48 patient derived glioblastoma cell cultures called the Human Glioma Cell Culture (HGCC) resource. We show that the HGCC cell cultures represent all transcriptional subtypes, carry genomic aberrations typical of glioblastoma, and initiate tumors in vivo. The HGCC is an open resource for translational glioblastoma research, made available through hgcc.se.In study 3 we extend the analysis of HGCC cell cultures both in terms of number (to over 100) and in terms of data types (adding mutation, methylation and drug response data). Large-scale drug profiling starting from over 1500 compounds identified two distinct groups of cell cultures defined by vulnerability to proteasome inhibition, p53/p21 activity, stemness and protein turnover. By applying machine learning methods to the combined drug profiling and matched genomics data we construct a first network of predictive biomarkers.In study 4 we use the methods developed in study 1 applied to the data generated in studies 2 and 3 to construct an integrative network model of HGCC and glioblastoma data from TCGA. We present an interactive method for exploring this network based on searching for network patterns representing specific hypotheses defined by the user.In conclusion, this thesis combines the development of integrative models with applications to novel data relevant for translational glioblastoma research. This work highlights several potentially therapeutically relevant aspects, and paves a path towards more comprehensive and informative models of glioblastoma.
|
|
| 4. |
- Kellgren, Therese, 1983-
(författare)
-
Hidden patterns that matter : statistical methods for analysis of DNA and RNA data
- 2020
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Understanding how the genetic variations can affect characteristics and function of organisms can help researchers and medical doctors to detect genetic alterations that cause disease and reveal genes that causes antibiotic resistance. The opportunities and progress associated with such data come however with challenges related to statistical analysis. It is only by using properly designed and employed tools, that we can extract the information about hidden patterns. In this thesis we present three types of such analysis. First, the genetic variant in the gene COL17A1 that causes corneal dystrophy with recurrent erosions is reveled. By studying Next-generation sequencing data, the order of the nucleotides in the DNAsequence was be obtained, which enabled us to detect interesting variants in the genome. Further, we present results of an experimental design study with the aim to make the best selection from a family that is affected by an inherited disease. In second part of the work, we analyzed a novel antibiotic resistance Staphylococcus epidermidis clone that is only found in northern Europe. By investigating its genetic data, we revealed similarities to a world known antibiotic resistance clone. As a result, the antibiotic resistance profile is established from the DNA sequences. Finally, we also focus on the challenges related to the abundance of genetic data from different sources. The increasing number of public gene expression datasets gives us opportunity to increase our understanding by using information from multiple sources simultaneously. Naturally, this requires merging independent datasets together. However, when doing so, the technical and biological variation in the joined data increases. We present a pre-processing method to construct gene co-expression networks from a large diverse gene-expression dataset.
|
|
